Well, like dagon observes, it's usually fast enough anyway - the performance bottleneck is usually elsewhere (disk access, process communication, network bandwidth...) and parse time is usually negligible in comparison. Zend Optimizer, for example, gets most of its performance boost from analysing and tuning the bytecode before caching it.
Scripts that would best benefit from it would be scripts that are both very large and very complex - unlike the usual database→webpage script (and as for the unusual ones, there's a good chance they'd benefit more from attention elsewhere).
But, like anything to do with optimisation, Your Results May Vary.